JP2000175038A - Image binary processing system on area basis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a top-down type segmentation method that finds out a photographing area on the basis of entire pixel connectivity. SOLUTION: An area-based binary processing system applies adaptive threshold processing and image rendering to a gray scale image to generate 1st and 2nd binary images. The gray scale image is sub-sampled to obtain a low resolution image and a position of a photographing image is detected in the low resolution image. Furthermore, a rectangular photographing image is identified in the detected photographing image and a classification map to distinguish pixels of the rectangular photographing image from other pixels is produced. Then a final binary image can be formed from the 1st and 2nd binary images on the basis of the classification map.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

FIELD OF THE INVENTION The present invention relates to a region-based binarization system that provides optimal binary image quality for mixed format documents.

[0002]

2. Description of the Related Art Magazine pages often have photographs printed in addition to text, line drawings and graphics.
When such a page is electronically captured by a scanner, the captured grayscale image (grey
scale image) is converted to an image
An onal representation is needed. Thus, image binarization techniques generally have two
There are two types. One is referred to as an adaptive thresholding technique, which is good for documents of the type that mainly include text and line art. The other is dither or error diffusion which reproduces gray shading in the form of a binary format.
fusion) technology. This is effective in binarizing a photographic image. For a mixed format document where text and photos are included in the captured document image, neither of the two binarization methods described above can produce satisfactory image quality of text and photos. A known solution to this problem is to segment the captured digital image into photographic and text regions so that a separate binarization process can be applied to each region for optimal image quality. That is.

A known segmentation method divides a mixed-format document into 4 × 4 blocks, classifies each block as text or image, and performs a short run of blocks.
run) to improve classification.
(See, for example, US Pat. No. 4,668,995 to Chen et al.)
See Issue 5). After the blocks of the image lines have been classified, a different binarization process is applied. Another known method is to use a run length for each scan line.
(run length) and extract rectangles from those run lengths
(rectangle), then classifying each rectangle into text or non-text, and finally merging the relevant text blocks into text areas (e.g., US to Cullen et al. See US Pat. No. 5,335,290).

[0004]

The two segmentation methods described above start by segmenting the information on a pixel-by-pixel or small block basis and then expand into regions. Is mentioned. These methods are not robust and are prone to classification errors, since the classification of text or non-text is based only on local information.

It is an object of the present invention to provide an overall pixel connectivity (g
To provide a top-down segmentation method for finding photographic regions based on global pixel connectivity) and to propose a region-based binarization system that obtains optimal binary image quality using the segmented results. is there.

[0006]

SUMMARY OF THE INVENTION The present invention is directed to a region-based binarization system that applies adaptive thresholding and image rendering, such as error diffusion (or dither), individually, and uses a grayscale system. Generating two binary images from the image, detecting the position of the photographic image in the low-resolution image, identifying the photographic image having a rectangular shape or boundary, and setting the photographic pixel to "1" and the non-photographic pixel to "0" And generating a classification bitmap to be marked as "" and assembling a final binary image from the two stored binary images based on the classification map.

[0007] The photo detection process uses global thresholding to convert low resolution grayscale images to two.
A binary image erosion process (binary imag) to remove the thin lines and the majority of characters.
e erosion process) and connected component analysis (c
applying an onnected component analysis to find objects, and using a size filter to eliminate small objects. The position of the large object is considered as the position of the photograph.

The present invention also includes converting the grayscale image to first and second binary images, detecting a position of the photographic image in the grayscale image, and detecting the position of the photographic image in the grayscale image. Identifying a photographic image having a rectangular boundary with, generating a classification map that distinguishes pixels in the photographic image having a rectangular boundary from the remaining pixels, and based on the classification map. Final 2 from binary image
Forming a value image.

[0009] The present invention further comprises the steps of capturing the image, locating the photographic image in the captured image, and identifying a photographic image having a rectangular boundary in the detected photographic image.
A region-based binarization process comprising: generating a classification map that distinguishes photographic pixels having rectangular boundaries from non-photographic pixels in the photographic image; and forming a final binary image based on the classification map. About.

The invention further comprises an image capture portion for capturing the image, a conversion portion for converting the captured image into digital image information representing the captured image, and processing the digital image information. Detecting the position of the photographic image in the captured image, identifying the photographic image having a rectangular boundary among the detected photographic images, and relocating the pixels in the photographic image having the rectangular boundary to the remaining pixels. A processing portion for generating a classification map that distinguishes from pixels.

[0011]

Referring to the drawings, wherein like reference numerals indicate like or corresponding parts throughout the different views, FIG. 1 is a block diagram of a region-based image binarization method. For digital grayscale image data as input, the above method works as follows.
Adaptive image thresholding (step 15a) is applied to convert the grayscale image (G) to a binary image (B) that exhibits good image quality of text and line art.
Convert to 1). An image rendering (step 15b) such as error diffusion or dither is applied to the same grayscale image (G) to obtain a rendered binary image (B2) exhibiting good image quality in the photographic portion of the image. . In step 15c,
Subsampling of the grayscale image is performed to provide a subsampled image (Gs). In step 16, the position of the rectangular photographic image in the sub-sampled image (Gs) is detected, while step 17
Detects a photographic image having a rectangular shape or boundary. At step 18, a classification map is generated that marks the rectangular photographic image with "1" and other pixels with "0", and the final binary image (B) is added to the generated classification map. Based two binary images B1 and B
2 is the result of image composition.
That is, assuming that the pixel at position (i, j) in the generated classification map is marked by "1" indicating a photographic pixel, the pixel at position (i, j) in image B2 is converted to binary image B. Is copied. on the other hand,
If the pixel at the position (i, j) in the classification map is “0” indicating a text pixel, the image B
The pixel at the middle position (i, j) is a copy of the binary image (B1).

FIG. 2 is a schematic diagram of an image capture assembly 300 for processing a captured image in accordance with the described features of the invention. The image capture assembly 300 includes an image capture portion 301 that captures the image, for example, in the form of a charge coupled device,
A conversion part 303 in the form of, for example, an A / D converter, for converting it into digital information representing the captured image
And a skier that includes The digital information is transmitted to an image processor 305 that processes the digital information in the manner described with respect to FIG. 1 and further described with respect to FIGS.

FIG. 3 shows the details of detecting a photographic image in a mixed format document (step 16). First, in the sub-sampling stage (15c in FIG. 1)
The grayscale image is subsampled every N pixels and every N scan lines to obtain a low resolution grayscale image (Gs). A fixed threshold value (global thresholding) (step 20) is provided,
The grayscale image (Gs) is converted into a binary image (Bs). A 3 × 3 binary erosion operation (step 21) is applied to all pixels of the binary image,
Remove thin lines and other fine objects, such as characters. The resulting image after the image erosion operation is saved as an image (Es). Connected component analysis (step 22) is applied to the image (Es) to group connected pixels. All groups of connected pixels are captured as objects. The boundary coordinates of an object define the position of the object. size·
Based on the filter (step 23), objects whose border size is greater than the size threshold value are considered photographs. As an example, the size filter may depend on the scan resolution. Illustrating Steps of the Method by Example FIG. 5 shows a scanned and printed page of a mixed format document page of a magazine containing text, lines, rectangular photographs and non-rectangular graphics (sunglasses graphics). Applying adaptive thresholding (step 15a, FIG. 1) to the grayscale image of FIG. 5 produces a binary image (B1). The binary image (B1) in FIG. 6 shows clear and vivid characters and lines, but the details of the shading of the photograph have been lost. By applying the error diffusion technique (step 15b in FIG. 1) to the same grayscale image, the resulting binary image (B2) shown in FIG. 7 retains the image details of the photographic area. Indicates that it is close to the actual photographic quality. However, the text image quality is blurred.
Comparing the two binary images (B1) and (B2) above, in order to generate a good binary image in a mixed format document, adaptive thresholding for text regions and error diffusion for photographic regions are considered. It is concluded that a combination is needed. In order to accomplish this task, the detection of a photographic area is required.

The detection process (step 16 in FIG. 1 and FIG. 3)
First, the grayscale image is sub-sampled to generate the small grayscale image (Gs) shown in FIG. Thereafter, the grayscale image (Gs) is thresholded with a fixed threshold value to generate a binary image (Bs). Result 2
The value image (Bs) is shown in FIG. As shown in FIG. 10, the binary image erosion operation (FIG. 3, step 21)
Applies to the image Es, where small letters and fine lines are removed and most of the remaining black pixels are in the photographic area. The bounding box of each object in the image (Es) is detected by a connected component (FIG. 3, step 22) that groups the connected black pixels of the binary image as individual objects (see, for example, U.S. Pat. 08 / 739,076). Eliminating the small objects (FIG. 3, step 23), the potential photographic bounding box 100 is shown in FIG. The boundary coordinates are converted to full resolution and
Is shown in The four border frames 100 are the positions of the detected photos. Not all of the photos in the detected border frame are necessarily rectangular.

The next step is to detect rectangular photographic objects within the positions of the four border frames 100.
(FIG. 1, step 17). This detection is performed by verifying all the characters in the boundary frame detected in the binary image (B1) in FIG. If there are any characters in the bounding box, the pictures in the bounding box are classified as non-rectangular. on the other hand,
If no characters are present in the border frame, the picture in the border frame is classified as a rectangular picture. In the above example, as shown in FIG. 13, characters are present in the border frame 100 'for the sunglasses graphics. Therefore, sunglasses graphics are considered non-rectangular photographs. Since the other three border frames do not contain characters, the photos they contain are classified as rectangular. Detection of photographic images with rectangular shapes or boundaries (Figure 1, step 1)
Details of 7) are shown in FIG. As shown in FIG. 4, a potential photographic image and a binary image (B1)
The information about the boundary coordinates is considered for connected component analysis (step 30). In step 30, a connected component analysis is performed to extract objects within each bounding box.
The largest object is a pictorial image (p
Note that small objects are classified as text or noise. In step 33, a check is made to determine if there are any small objects (characters) that are not geometrically located outside the boundaries of the largest object. If the answer to step 33 is YES,
The object is a photographic image with non-rectangular boundaries. If the answer to step 33 is NO, the object is a photographic image with rectangular boundaries.

Next, a classification map (FIG. 1, step 18) is generated by filling the boundaries of the three rectangular photographs with black pixels as shown in FIG. The final binary image (B) is assembled from the binary images (B1) and (B2) based on the classification map. Pixels of image (B) are copies of image (B1) in the text area of the classification map, and images (B) are copies of image (B2) in the photographic area (black area of the classification map). The result is shown in FIG.
It is shown in FIG.

[Brief description of the drawings]

FIG. 1 is a flowchart showing the steps of an area-based binarization system for a mixed format document.

FIG. 2 is a schematic diagram of an apparatus according to the present invention.

FIG. 3 is a flowchart outlining each step of the photo detection process.

FIG. 4 is a flowchart outlining steps for detecting a photographic image having a rectangular shape or boundary.

FIG. 5 is an example of a grayscale blended document (G) printed digitally.

FIG. 6 is a thresholded image (B1) of the image (G) of FIG. 3 using the adaptive thresholding method.

FIG. 7 is a thresholded image (B2) of the image (G) of FIG. 3 using the error diffusion method.

FIG. 8 is a sub-sampled image (Gs) of the image (G) of FIG. 3 using 4: 1 size reduction.

9 is a thresholded image (Bs) of the image of FIG. 6 (Gs) using fixed thresholding.

FIG. 10 is a resultant binary image (Es) of the image erosion process on the image (Bs) of FIG. 9;

11 is a boundary frame of a large object detected for the binary image (E) in FIG.

FIG. 12 is a photograph position detected in the image (B1) of FIG. 6;

FIG. 13 is an example of a border frame including text for a non-rectangular photograph.

FIG. 14 is a bitmap showing an area of a detected rectangular photograph.

FIG. 15 is a final binary image using the method of the present invention.

[Explanation of symbols]

 100: Potential photo border 300: Image capture assembly 301: Image capture 303: A / D converter 305: Image processor

Claims (9)

[Claims] 1. converting a grayscale image into first and second binary images; detecting a position of the photographic image in the grayscale image; Identifying a photographic image having a rectangular boundary within; generating a classification map that distinguishes pixels in the photographic image having the rectangular boundary from the remaining pixels; and, based on the classification map, the first and second images. Forming a final binary image from the two binary images. 2. The method of claim 1, further comprising the step of subsampling the grayscale image to obtain a subsampled image, wherein the detecting step comprises detecting a position of a photographic image in the subsampled image. The method of claim 1. 3. The method of claim 1, wherein the grayscale image has been captured from a document that includes at least a photographic portion and a text portion. 4. The method of claim 1, wherein the converting step comprises:
Applying one of the first and second binary images by applying an adaptive thresholding technique to the image; and applying the image rendering technique to the first and second grayscale images. Acquiring the other of the second binary image. 5. The method of claim 4, wherein said image rendering technique comprises an error diffusion process. 6. The method of claim 4, wherein said image rendering technique comprises a dither process. 7. The method of claim 2, wherein said sub-sampled image is a low resolution image. 8. The detecting step further comprises: converting the sub-sampled image into a further binary image; removing thin lines and characters from the further binary image; Performing connected component analysis on the further binary image to group the connected pixels in, wherein the group of connected pixels is identified as an object in the further binary image; and 3. The method of claim 2, comprising: designating an object having a size greater than the threshold value as a photographic image at. 9. The method according to claim 8, wherein the specifying step includes the step of specifying an object larger than the threshold value as an object image and using a size filter to exclude objects smaller than the threshold value.